Fusion of Radical Vectors and Capsule Network Legal Named Entity Recognition

Abstract

The present invention provides a word segmentation and capsule network legal named entity recognition method that integrates radical vectors, which is characterized in that it specifically includes the following steps: S1: word segmentation; S2: constructing legal data sets; S3: extracting inter-sentence features; S4: Extract the internal feature H of the sentence; S5: jointly represent the internal sentence feature and the inter-sentence feature of the corpus. Through the technical solution of the present invention, the characterization ability of the word vector is improved, so that the feature extraction stage does not require a complicated network, and the structure can obtain rich context information, which improves the word segmentation effect on the small-scale domain labeling data set, and helps NER tasks. Data preprocessing, find accurate entity boundaries, enhance the adaptability of the model in the legal field, and achieve high-precision named entity recognition results in the legal field with special entity representations and word formation methods. Named entity recognition in the legal field can play an important role in intelligent justice, helping to extract key information in texts.

Description

technical field [0001] The present invention relates to the technical field of natural language processing, in particular to a method for recognizing a legal named entity by combining radical vectors for word segmentation and capsule network. Background technique [0002] For the Chinese NER task, we mainly face the following problems: (1) Entity boundary problem, one aspect of judging whether an entity is correctly recognized is whether the entity boundary is correct. The reason that affects the recognition of entity boundaries is that Chinese is different from English. There are no obvious marker words and spaces in Chinese sequences to distinguish word boundaries. For example, English entities can be judged based on whether the first letter is capitalized. The first step in the NER task is to determine the boundary of words , according to certain rules to cut into words one by one. Chinese word segmentation and NER tasks affect each other, and the accuracy of word segmen...

AI Technical Summary

Problems solved by technology

In addition, compared with other tasks, NER has less training corpus, which leads to poor performance of general-purpose NER models that recognize multiple types.

For domain NER, the limitation of the size of the training corpus also leads to poor experimental results

Images